sagieppel/Focusing-attention-of-Fully-convolutional-neural-networks-on-Region-of-interest-ROI-input-map-

This project contains code for a fully convolutional neural network (FCN) for semantic segmentation with a region of interest (ROI) map as an additional input (Figure 1). The net receives image and ROI as a binary map with pixels corresponding to ROI marked 1, and produce pixel-wise annotation of the ROI region of the image. The method is based on using region selective features This code was tested on for semantic segmentation task of materials in transparent vessels where the vessel area of the image was set as the ROI.

Python